Release review of 0.12.0 #634

Merged
lilleman merged 14 commits from release-review-0.12.0 into main 2026-11-19 14:12:13 +01:00
Owner

The release review of 0.12.0: the architecture and principles passes over the whole project, and their fixes through stability, security and prose.

  • Architecture, 4 rounds, ends APPROVE WITH NITS. What loop-claude and do-epic share is now bin/chunk_driver/, so do-epic no longer imports loop-claude's internals. Review-state reading is one module, bin/forge/reviews.py, and Jira's client sits in bin/atlassian/. A repo check keeps the package layers from importing upwards.
  • Principles, 2 rounds. No Python file patches a global any more: clocks, sleeps and the environment are arguments, and a repo check holds that. One 🔴: the decision to keep traces on the machine contradicted the traces principle. I scoped the principle to say a tool that runs only on its user's machine archives its trace there, as the 2026-11-16 ruling in that decision reads. This is my choice among the auditor's options; say if the decision should change instead.
  • Stability, 1 round, 4 🟡 fixed. Prose, 1 round, 11 🟡 fixed. Security: no finding.
  • Comprehension check: paired ruling better at 914b8dc2 against aedc5ed3. Mid A and the inherited architect both voted Navigation, Locality and Shape better, Self-sufficiency same.

Declined or left:

  • Grouping Cycle's fields: the decision its docstring cites keeps them flat.
  • About 26 hard places the review doesn't touch belong to floor item 300.
  • The principles ❓ on the note about 0.10.1's semver breach: "Semantic versioning, read strictly" already says an exception doesn't set a rule.
  • Effects left with defaults on purpose: the notifier's http_json(request=…), todo_item.main(plan_of=…), review_replay.main(scratch=None), and adf_codec's machine_slot() and docker().
  • Mutants left alive on changed lines: chunk_pr/main.py:142, which needs sleep handed into step; review_replay/replay.py:97's ignore_cleanup_errors; and equivalents on encodings and timeouts. The run after the new tests was lost with a session and not repeated.

Skipped: the overrides pass, since rule-hook overrides lists none; the product-owner pass, since no external surface changed; an outside auto-reviewer, since this repo has none.

The release review of 0.12.0: the architecture and principles passes over the whole project, and their fixes through stability, security and prose. - **Architecture**, 4 rounds, ends APPROVE WITH NITS. What loop-claude and do-epic share is now `bin/chunk_driver/`, so do-epic no longer imports loop-claude's internals. Review-state reading is one module, `bin/forge/reviews.py`, and Jira's client sits in `bin/atlassian/`. A repo check keeps the package layers from importing upwards. - **Principles**, 2 rounds. No Python file patches a global any more: clocks, sleeps and the environment are arguments, and a repo check holds that. One 🔴: the decision to keep traces on the machine contradicted the traces principle. I scoped the principle to say a tool that runs only on its user's machine archives its trace there, as the 2026-11-16 ruling in that decision reads. This is my choice among the auditor's options; say if the decision should change instead. - **Stability**, 1 round, 4 🟡 fixed. **Prose**, 1 round, 11 🟡 fixed. **Security**: no finding. - **Comprehension check**: paired ruling **better** at 914b8dc2 against aedc5ed3. Mid A and the inherited architect both voted Navigation, Locality and Shape better, Self-sufficiency same. Declined or left: - Grouping `Cycle`'s fields: the decision its docstring cites keeps them flat. - About 26 hard places the review doesn't touch belong to floor item 300. - The principles ❓ on the note about 0.10.1's semver breach: "Semantic versioning, read strictly" already says an exception doesn't set a rule. - Effects left with defaults on purpose: the notifier's `http_json(request=…)`, `todo_item.main(plan_of=…)`, `review_replay.main(scratch=None)`, and adf_codec's `machine_slot()` and `docker()`. - Mutants left alive on changed lines: `chunk_pr/main.py:142`, which needs `sleep` handed into `step`; `review_replay/replay.py:97`'s `ignore_cleanup_errors`; and equivalents on encodings and timeouts. The run after the new tests was lost with a session and not repeated. Skipped: the overrides pass, since `rule-hook overrides` lists none; the product-owner pass, since no external surface changed; an outside auto-reviewer, since this repo has none.
lilleman added 2 commits 2026-11-19 10:34:15 +01:00
lilleman added 1 commit 2026-11-19 11:14:48 +01:00
lilleman added 1 commit 2026-11-19 11:29:36 +01:00
lilleman added 1 commit 2026-11-19 11:43:43 +01:00
lilleman added 1 commit 2026-11-19 11:54:39 +01:00
lilleman added 1 commit 2026-11-19 12:42:08 +01:00
lilleman added 1 commit 2026-11-19 12:55:39 +01:00
lilleman added 1 commit 2026-11-19 13:09:59 +01:00
Answer security round 1: 0 🔴, 0 🟡 (0 misread); nothing to fix
CI / tests (pull_request) Successful in 3m51s
2edbde8071
lilleman added 2 commits 2026-11-19 13:44:25 +01:00
lilleman added 1 commit 2026-11-19 14:04:16 +01:00
lilleman marked the pull request as ready for review 2026-11-19 14:04:30 +01:00
lilleman added 2 commits 2026-11-19 14:08:20 +01:00
lilleman merged commit a7e1297805 into main 2026-11-19 14:12:13 +01:00
lilleman deleted branch release-review-0.12.0 2026-11-19 14:12:13 +01:00
Sign in to join this conversation.
No Reviewers
No Label
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: lilleman/claude-config#634